Product1 publisher3 min readPublished
Amodei asks Beijing to pace the frontier while Washington widens its lead
Dario Amodei made global pacing conditional on democracies keeping as large a lead as possible over China, and the officials and analysts who answered him in Beijing objected to the terms. Beijing rejected those terms in public.
The Product Desk · Product desk
What happened
- Anthropic CEO Dario Amodei published a post on Saturday titled "We Must Pace the Frontier," arguing that global pacing will require cooperation with China, which he called the autocratic country with by far the most advanced AI capabilities.
- Guo Jiakun, a spokesperson for China's Ministry of Foreign Affairs, was asked about the post at a Monday press conference and called it "fear-mongering, confrontation, and vicious competition" that would hamper global AI cooperation.
- Wired reports that China's government and Chinese AI companies are deeply skeptical of the proposal because they think they are being offered a bad deal, not because they dismiss the risks of the technology.
- Amodei later admitted in a television interview that he has "no idea how to negotiate with China."
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint No Chinese party has agreed to pace anything, so a plan that defers capability-dependent work until the frontier settles is waiting indefinitely.
- decision Teams have to choose in the open now: fund the work that only pays off if models keep improving, or state in the plan that they are betting on a pause Beijing has publicly refused on these terms.
- contradiction Wired attributes the rejection to the deal's terms, while CNAS's Michelle Nie says Beijing doubts AGI is achievable at all; if the doubt is about the premise, improving the terms would not produce agreement.
For the person who has to defend a capability plan on Friday, the first thing to check in a pacing proposal is who has to sign it.
The essay asks for two different kinds of thing. A crackdown on chip smuggling and a block on unauthorized model distillation by Chinese companies are steps Washington can take on its own, and Amodei urged both [2]. Pacing the frontier is the step that needs Beijing [1]. He made the second conditional on the first, writing that a "key part" of the strategy is keeping the lead held by the US and other democracies "as large as possible" [2]. Washington can start the enforcement half without Beijing [14].
That ordering is what the China-focused readers in Wired's account fixed on. "His essay specifically says the pacing of AI is conditional to increasing the lead over China," said Alvin Wang Graylin, a senior fellow at the Asia Society Policy Institute's Center for China Analysis. That, he said, "makes the proposal feel inauthentic to Chinese readers" [7].
"The gulf lies between an Anthropic-led vision that would only subjugate and restrict China first to maximize leverage before talking," said Kevin Xu, founder of the technology hedge fund Interconnected Capital and a former White House staffer under President Barack Obama. "Meanwhile, China would only talk if it's being treated from a position of equals" [6]. Xu also said the AI safety community in China is growing and "shares many of the same concerns that American labs do" [5].
The responses Wired collected do accept that AI carries risk [15]. The foreign ministry said the day after the press conference that "China puts equal emphasis on development and security in terms of AI" [4]. China's minister of state security, Chen Yixin, warned in a recent essay that AI could fuel political disinformation, cyberattacks, espionage, social unrest and new military threats [12]. Michelle Nie, a visiting fellow at the Center for a New American Security, said Beijing "is starting to recognize loss-of-control risks," while the risk it seems most worried about is the threat AI could pose to CCP rule [10].
Wired's account is built from public statements and named analysts' assessments, and it does not report any lab moving a release date [16]. It also notes that the Silicon Valley executives who spent years arguing the US is locked in an AI arms race with China spent the past week saying the two countries urgently need to work together to slow the race down [13].
Every roadmap item carries an assumption about how the models change over the next two quarters, and someone has to hold that assumption for the item to work. Items that assume steady improvement depend on your vendor's release cadence. Items that assume the frontier holds still depend on two governments agreeing terms, and the terms on offer this week arrived with a condition attached [2].
What to watch
- Whether Anthropic revises the pacing proposal to drop the condition on widening the US lead that Graylin flagged.
- Whether Beijing or a Chinese lab names terms it would accept for talks; that would give planners a counterparty.
- Whether US enforcement on chip smuggling and model distillation tightens, since Washington can act on that half alone.